WIP: Power VSX support almost completed

This commit is contained in:
Apostolos Tapsas
2021-10-14 10:33:10 +00:00
committed by apostolos
parent 28f8f30866
commit 3423ea5b2b
3 changed files with 254 additions and 381 deletions

View File

@@ -1,5 +1,6 @@
/*
* Copyright (c) 2015-2020, Intel Corporation
* Copyright (c) 2020-2021, VectorCamp PC
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
@@ -51,8 +52,8 @@ typedef __vector int16_t int16x8_t;
typedef __vector uint8_t uint8x16_t;
typedef __vector int8_t int8x16_t;
static really_inline m128 ones128(void) {
// the value in function must be a signed literal in range -16 to 15
return (m128) vec_splat_s8(-1);
}
@@ -80,14 +81,15 @@ static really_inline int isnonzero128(m128 a) {
* mask indicating which 32-bit words contain differences.
*/
static really_inline u32 diffrich128(m128 a, m128 b) {
static const m128 movemask = { 1, 2, 4, 8 };
m128 mask = (m128) vec_cmpeq(a, b);
mask = vec_and(vec_xor(mask, mask), movemask);
m128 sum = vec_sums(mask, zeroes128());
sum = vec_sld(zeroes128(), sum, 4);
s32 ALIGN_ATTR(16) x;
vec_ste(sum, 0, &x);
return x;
static const m128 movemask = { 1, 2, 4, 8 };
m128 mask = (m128) vec_cmpeq(a, b); // _mm_cmpeq_epi32 (a, b);
mask = vec_and(not128(mask), movemask);
m128 sum = vec_sums(mask, zeroes128());
//sum = vec_sld(zeroes128(), sum, 4);
//s32 ALIGN_ATTR(16) x;
//vec_ste(sum, 0, &x);
//return x; // it could be ~(movemask_128(mask)) & 0x;
return sum[3];
}
/**
@@ -97,12 +99,13 @@ static really_inline u32 diffrich128(m128 a, m128 b) {
static really_inline u32 diffrich64_128(m128 a, m128 b) {
static const uint64x2_t movemask = { 1, 4 };
uint64x2_t mask = (uint64x2_t) vec_cmpeq((uint64x2_t)a, (uint64x2_t)b);
mask = vec_and(vec_xor(mask, mask), movemask);
mask = (uint64x2_t) vec_and((uint64x2_t)not128((m128)mask), movemask);
m128 sum = vec_sums((m128)mask, zeroes128());
sum = vec_sld(zeroes128(), sum, 4);
s32 ALIGN_ATTR(16) x;
vec_ste(sum, 0, &x);
return x;
//sum = vec_sld(zeroes128(), sum, 4);
//s32 ALIGN_ATTR(16) x;
//vec_ste(sum, 0, &x);
//return x;
return sum[3];
}
static really_really_inline
@@ -116,32 +119,59 @@ m128 sub_2x64(m128 a, m128 b) {
}
static really_really_inline
m128 lshift_m128(m128 a, unsigned UNUSED b) {
// #warning FIXME
// b must be 4 bit literal
return (m128) vec_sld(a, zeroes128(), 0);
m128 lshift_m128(m128 a, unsigned b) {
switch(b){
case 1: return vec_sld(a, zeroes128(), 1); break;
case 2: return vec_sld(a, zeroes128(), 2); break;
case 3: return vec_sld(a, zeroes128(), 3); break;
case 4: return vec_sld(a, zeroes128(), 4); break;
case 5: return vec_sld(a, zeroes128(), 5); break;
case 6: return vec_sld(a, zeroes128(), 6); break;
case 7: return vec_sld(a, zeroes128(), 7); break;
case 8: return vec_sld(a, zeroes128(), 8); break;
case 9: return vec_sld(a, zeroes128(), 9); break;
case 10: return vec_sld(a, zeroes128(), 10); break;
case 11: return vec_sld(a, zeroes128(), 11); break;
case 12: return vec_sld(a, zeroes128(), 12); break;
case 13: return vec_sld(a, zeroes128(), 13); break;
case 14: return vec_sld(a, zeroes128(), 14); break;
case 15: return vec_sld(a, zeroes128(), 15); break;
}
return a;
}
static really_really_inline
m128 rshift_m128(m128 a, unsigned UNUSED b) {
// #warning FIXME
// b must be 4 bit literal
return (m128) vec_sld(zeroes128(), a, 0 - 0);
m128 rshift_m128(m128 a, unsigned b) {
switch(b){
case 1: return vec_sld(zeroes128(), a, 15); break;
case 2: return vec_sld(zeroes128(), a, 14); break;
case 3: return vec_sld(zeroes128(), a, 13); break;
case 4: return vec_sld(zeroes128(), a, 12); break;
case 5: return vec_sld(zeroes128(), a, 11); break;
case 6: return vec_sld(zeroes128(), a, 10); break;
case 7: return vec_sld(zeroes128(), a, 9); break;
case 8: return vec_sld(zeroes128(), a, 8); break;
case 9: return vec_sld(zeroes128(), a, 7); break;
case 10: return vec_sld(zeroes128(), a, 6); break;
case 11: return vec_sld(zeroes128(), a, 5); break;
case 12: return vec_sld(zeroes128(), a, 4); break;
case 13: return vec_sld(zeroes128(), a, 3); break;
case 14: return vec_sld(zeroes128(), a, 2); break;
case 15: return vec_sld(zeroes128(), a, 1); break;
}
return a;
}
static really_really_inline
m128 lshift64_m128(m128 a, unsigned UNUSED b) {
// #warnint FIXME
// b must be 4 bit literal
return (m128) vec_sld(zeroes128(), a, 0);
m128 lshift64_m128(m128 a, unsigned b) {
uint64x2_t shift_indices = vec_splats((uint64_t)b);
return (m128) vec_sl((int64x2_t)a, shift_indices);
}
static really_really_inline
m128 rshift64_m128(m128 a, unsigned UNUSED b) {
// warnint FIXME
// b must be 4 bit literal
return (m128) vec_sld(zeroes128(), a, 0);
m128 rshift64_m128(m128 a, unsigned b) {
uint64x2_t shift_indices = vec_splats((uint64_t)b);
return (m128) vec_sr((int64x2_t)a, shift_indices);
}
static really_inline m128 eq128(m128 a, m128 b) {
@@ -149,39 +179,36 @@ static really_inline m128 eq128(m128 a, m128 b) {
}
static really_inline m128 eq64_m128(m128 a, m128 b) {
return (m128) vec_cmpeq((uint64x2_t)a, (uint64x2_t)b);
return (m128) vec_cmpeq((uint64x2_t)a, (uint64x2_t)b);
}
static really_inline u32 movemask128(m128 a) {
//static const uint8x16_t powers = { 1, 2, 4, 8, 16, 32, 64, 128, 1, 2, 4, 8, 16, 32, 64, 128 };
uint8x16_t s1 = vec_sr((uint8x16_t)a, vec_splat_u8(7));
uint16x8_t ss = vec_sr((uint16x8_t)s1, vec_splat_u16(7));
uint16x8_t res_and = vec_and((uint16x8_t)s1, vec_splats((uint16_t)0xff));
uint16x8_t s2 = vec_or((uint16x8_t)ss, res_and);
// Compute the mask from the input
//uint64x2_t mask = vpaddlq_u32(vpaddlq_u16(vpaddlq_u8(vandq_u8((uint8x16_t)a, powers))));
//uint64x2_t mask1 = (m128)vextq_s8(mask, zeroes128(), 7);
//mask = vorrq_u8(mask, mask1);
uint32x4_t ss2 = vec_sr((uint32x4_t)s2, vec_splat_u32(14));
uint32x4_t res_and2 = vec_and((uint32x4_t)s2, vec_splats((uint32_t)0xff));
uint32x4_t s3 = vec_or((uint32x4_t)ss2, res_and2);
// Get the resulting bytes
//uint16_t output;
//vst1q_lane_u16((uint16_t*)&output, (uint16x8_t)mask, 0);
//return output;
// #warning FIXME
return !!diff128(a, zeroes128());
uint64x2_t ss3 = vec_sr((uint64x2_t)s3, (uint64x2_t)vec_splats(28));
uint64x2_t res_and3 = vec_and((uint64x2_t)s3, vec_splats((uint64_t)0xff));
uint64x2_t s4 = vec_or((uint64x2_t)ss3, res_and3);
uint64x2_t ss4 = vec_sld((uint64x2_t)vec_splats(0), s4, 9);
uint64x2_t res_and4 = vec_and((uint64x2_t)ss4, vec_splats((uint64_t)0xff));
uint64x2_t s5 = vec_or((uint64x2_t)ss4, res_and4);
return s5[0];
}
static really_inline m128 set1_16x8(u8 UNUSED c) {
// warning FIXME
// c must be 5 bit literal
// a solution is to use vec_splats
//return (m128) vec_splat_u8(0);
static really_inline m128 set1_16x8(u8 c) {
return (m128) vec_splats(c);
}
static really_inline m128 set1_4x32(u32 UNUSED c) {
// warning FIXME
// c must be 5 bit literal
// a solution is to use vec_splats
// return (m128) vec_splat_u32(0);
static really_inline m128 set1_4x32(u32 c) {
return (m128) vec_splats(c);
}
@@ -196,15 +223,15 @@ static really_inline u32 movd(const m128 in) {
}
static really_inline u64a movq(const m128 in) {
//return vgetq_lane_u64((uint64x2_t) in, 0);
return !!diff128(in, zeroes128());
// #warning FIXME
u64a ALIGN_ATTR(16) a[2];
vec_xst((uint64x2_t) in, 0, a);
return a[0];
}
/* another form of movq */
static really_inline
m128 load_m128_from_u64a(const u64a *p) {
return (m128) vec_ld(0,p);
return (m128) vec_ld(0, p);
}
@@ -236,8 +263,8 @@ static really_inline u32 extract32from128(const m128 in, unsigned imm) {
return vec_any_ne(in,lshift_m128(in,imm));
}
static really_inline u64a extract64from128(const m128 in, unsigned imm) {
/*
static really_inline u64a extract64from128(const m128 UNUSED in, unsigned UNUSED imm) {
/* is this
#if defined(HS_OPTIMIZE)
return vgetq_lane_u64((uint64x2_t) in, imm);
#else
@@ -253,21 +280,32 @@ static really_inline u64a extract64from128(const m128 in, unsigned imm) {
break;
}
#endif
*/
// #warning FIXME
return vec_any_ne(in,lshift_m128(in,imm));
*/
/*
u64a ALIGN_ATTR(16) a[2];
vec_xst((uint64x2_t) in, 0, a);
switch(imm) {
case 0: return a[0]; break;
case 1: return a[1]; break;
default: return 0; break;
}
*/
return 0;
}
static really_inline m128 low64from128(const m128 in) {
//return vcombine_u64(vget_low_u64(in), vdup_n_u64(0));
// #warning FIXME
return in;
//u64a ALIGN_ATTR(16) a[2];
//vec_xst((uint64x2_t) in, 0, a);
//return a[1];
return vec_add(in, in);
}
static really_inline m128 high64from128(const m128 in) {
//return vcombine_u64(vget_high_u64(in), vdup_n_u64(0));
// #warning FIXME
return in;
//u64a ALIGN_ATTR(16) a[2];
//vec_xst((uint64x2_t) in, 0, a);
//return a[0];
return vec_add(in, in);
}
@@ -288,37 +326,29 @@ static really_inline m128 or128(m128 a, m128 b) {
}
static really_inline m128 andnot128(m128 a, m128 b) {
m128 and_res = and128(a,b);
return (m128) not128(and_res);
// or
//return (m128) not128(and128(a,b));
return (m128) and128(not128(a),b);
}
// aligned load
static really_inline m128 load128(const void *ptr) {
assert(ISALIGNED_N(ptr, alignof(m128)));
//return (m128) vec_ld(0, ptr);
// #warning FIXME
return zeroes128();
return (m128) vec_xl(0, (const int32_t*)ptr);
}
// aligned store
static really_inline void store128(void *ptr, m128 UNUSED a) {
static really_inline void store128(void *ptr, m128 a) {
assert(ISALIGNED_N(ptr, alignof(m128)));
//vec_st(a, 0, ptr);
// warning FIXME
vec_st(a, 0, (int32_t*)ptr);
}
// unaligned load
static really_inline m128 loadu128(const void UNUSED *ptr) {
//return (m128) vec_ld(0, ptr);
// #warning FIXME
return zeroes128();
static really_inline m128 loadu128(const void *ptr) {
return (m128) vec_xl(0, (const int64_t*)ptr);
}
// unaligned store
static really_inline void storeu128(void UNUSED *ptr, m128 UNUSED a) {
// #warning FIXME
static really_inline void storeu128(void *ptr, m128 a) {
vec_st(a, 0, (int32_t*)ptr);
}
// packed unaligned store of first N bytes
@@ -338,11 +368,10 @@ m128 loadbytes128(const void *ptr, unsigned int n) {
}
//#define CASE_ALIGN_VECTORS(a, b, offset) case offset: return (m128)vextq_s8((int8x16_t)(a), (int8x16_t)(b), (offset)); break;
#define CASE_ALIGN_VECTORS(a, b, offset) case offset: return (m128)vec_sld((int8x16_t)(b), (int8x16_t)(a), (16 - offset)); break;
static really_really_inline
m128 palignr_imm(m128 r, m128 l, int offset) {
/*
switch (offset) {
case 0: return l; break;
CASE_ALIGN_VECTORS(l, r, 1);
@@ -361,56 +390,39 @@ m128 palignr_imm(m128 r, m128 l, int offset) {
CASE_ALIGN_VECTORS(l, r, 14);
CASE_ALIGN_VECTORS(l, r, 15);
case 16: return r; break;
default:
return zeroes128();
break;
}
*/
// #warning FIXME
return (m128) vec_cmpeq(r,lshift_m128(l,offset));
default: return zeroes128(); break;
}
}
static really_really_inline
m128 palignr(m128 r, m128 l, int offset) {
/*
#if defined(HS_OPTIMIZE)
return (m128)vextq_s8((int8x16_t)l, (int8x16_t)r, offset);
return (m128)vec_sld((int8x16_t)l, (int8x16_t)r, offset);
#else
return palignr_imm(r, l, offset);
#endif
*/
// #warning FIXME
return (m128) vec_cmpeq(r, lshift_m128(l,offset));
}
#undef CASE_ALIGN_VECTORS
static really_really_inline
m128 rshiftbyte_m128(m128 a, unsigned b) {
// #warning FIXME
// return vec_sro(a,b);
return rshift_m128(a,b);
}
static really_really_inline
m128 lshiftbyte_m128(m128 a, unsigned b) {
//#warning FIXME
//return vec_slo(a,b);
return lshift_m128(a,b);
return lshift_m128(a,b);
}
static really_inline
m128 variable_byte_shift_m128(m128 in, s32 amount) {
/*
assert(amount >= -16 && amount <= 16);
static const uint8x16_t vbs_mask = { 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f };
const uint8x16_t outside_mask = set1_16x8(0xf0);
m128 shift_mask = palignr_imm(vbs_mask, outside_mask, 16 - amount);
return vqtbl1q_s8(in, shift_mask);
*/
// #warning FIXME
return lshift_m128(in,amount);
if (amount < 0){
return palignr_imm(zeroes128(), in, -amount);
} else{
return palignr_imm(in, zeroes128(), 16 - amount);
}
}
#ifdef __cplusplus
@@ -450,28 +462,22 @@ char testbit128(m128 val, unsigned int n) {
static really_inline
m128 pshufb_m128(m128 a, m128 b) {
/* On Intel, if bit 0x80 is set, then result is zero, otherwise which the lane it is &0xf.
In NEON, if >=16, then the result is zero, otherwise it is that lane.
btranslated is the version that is converted from Intel to NEON. */
//int8x16_t btranslated = vandq_s8((int8x16_t)b,vdupq_n_s8(0x8f));
//return (m128)vqtbl1q_s8((int8x16_t)a, (uint8x16_t)btranslated);
// #warning FIXME
return (m128) vec_max((int8x16_t)a, (int8x16_t)b);
return (m128) vec_permxor((int8x16_t)vec_splat_s8(0), (int8x16_t)a, (int8x16_t)b);
}
static really_inline
m128 max_u8_m128(m128 a, m128 b) {
return (m128) vec_max((int8x16_t)a, (int8x16_t)b);
return (m128) vec_max((uint8x16_t)a, (uint8x16_t)b);
}
static really_inline
m128 min_u8_m128(m128 a, m128 b) {
return (m128) vec_min((int8x16_t)a, (int8x16_t)b);
return (m128) vec_min((uint8x16_t)a, (uint8x16_t)b);
}
static really_inline
m128 sadd_u8_m128(m128 a, m128 b) {
return (m128) vec_add((uint8x16_t)a, (uint8x16_t)b);
return (m128) vec_adds((uint8x16_t)a, (uint8x16_t)b);
}
static really_inline
@@ -480,19 +486,15 @@ m128 sub_u8_m128(m128 a, m128 b) {
}
static really_inline
m128 set4x32(u32 UNUSED x3, u32 UNUSED x2, u32 UNUSED x1, u32 UNUSED x0) {
//uint32_t ALIGN_ATTR(16) data[4] = { x0, x1, x2, x3 };
//return (m128) vec_splat_u32(data);
// #warning FIXME
return zeroes128();
m128 set4x32(u32 x3, u32 x2, u32 x1, u32 x0) {
uint32x4_t v = { x0, x1, x2, x3 };
return (m128) v;
}
static really_inline
m128 set2x64(u64a UNUSED hi, u64a UNUSED lo) {
//uint64_t ALIGN_ATTR(16) data[2] = { lo, hi };
//return (m128) vec_splats(data);
// #warning FIXME
return zeroes128();
m128 set2x64(u64a hi, u64a lo) {
uint64x2_t v = { lo, hi };
return (m128) v;
}
#endif // ARCH_PPC64EL_SIMD_UTILS_H